In this analysis, psychosis spectrum (PS) status is predicted based on 22q11.2 deletion syndrome (22q) characteristics.
First, PCA is performed on the 22q sample to estimate the relative contributions, i.e. the loadings, of each gestalt score to the axes of greatest variability (PCs) in this group.
These loadings are then used to compute “22q-like” factor scores for the NC and PS groups: ROC analysis shows PC2 alone correctly predicts PS status 67.8% of the time.
Additionally, Levene’s test shows that variance from the schizophrenia (SZ) group differs from that of clinical risk (CR) on PC2.
Sample sizes for face2gene scores and the original prediction task:
| 22q | PS | TD |
|---|---|---|
| 150 | 55 | 93 |
Total sample sizes in 03/08/22 data (first row) and non-missing data per variable:
| 22q | PS | TD | total | |
|---|---|---|---|---|
| 150 | 55 | 93 | 298 | |
| height | 50 | 19 | 14 | 83 |
| weight | 110 | 36 | 29 | 175 |
| GAF_C | 128 | 29 | 23 | 180 |
| GAF_H | 125 | 29 | 23 | 177 |
| MMSE | 111 | 41 | 41 | 193 |
| VIQ | 76 | 0 | 0 | 76 |
| PIQ | 76 | 0 | 0 | 76 |
| FSIQ | 76 | 0 | 0 | 76 |
| DSM Dx | 150 | 0 | 0 | 150 |
| PS | 148 | 0 | 0 | 148 |
Note: Emotrics available for all records.
PCA was performed without scaling on the 22q sample alone.
Using the “elbow rule”, subsequent analysis will focus on the first 4 PCs.
Loadings for the first 2 PCs are plotted below.
## Analysis of Variance Table
##
## Response: PC2
## Df Sum Sq Mean Sq F value Pr(>F)
## sex 1 0.2870 0.28702 4.9192 0.028207 *
## age 1 0.6551 0.65507 11.2270 0.001042 **
## race 4 0.3486 0.08715 1.4935 0.207518
## Residuals 137 7.9937 0.05835
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## Warning: Removed 6 rows containing non-finite values (stat_smooth).
## Warning: Removed 6 rows containing missing values (geom_point).
PC2 scores were correlated with Brow Height (Left and Right), Marginal Reflex Distance 2 (Left and Right), and Philtrum measurements.
| PC1 | p | PC2 | p | |
|---|---|---|---|---|
| Brow_Height_Right | -0.1837632 | 0.0243851 | -0.2986967 | 0.0002050 |
| Brow_Height_Left | -0.1143783 | 0.1634075 | -0.3526283 | 0.0000096 |
| Marginal_Reflex_Distance_1_Right | 0.0314815 | 0.7021367 | -0.1502111 | 0.0665447 |
| Marginal_Reflex_Distance_1_Left | 0.1122420 | 0.1714674 | -0.1216564 | 0.1380664 |
| Marginal_Reflex_Distance_2_Right | 0.1104928 | 0.1782834 | -0.1824594 | 0.0254329 |
| Marginal_Reflex_Distance_2_Left | 0.1277416 | 0.1192784 | -0.1826827 | 0.0252508 |
| Philtrum | -0.0099248 | 0.9040557 | -0.3024779 | 0.0001685 |
There were no significant relationships with IQ.
| PC1 | p | PC2 | p | |
|---|---|---|---|---|
| VIQ | 0.0121797 | 0.9168325 | -0.1635336 | 0.1580832 |
| PIQ | -0.0371284 | 0.7501639 | -0.1593620 | 0.1691110 |
| FSIQ | 0.2568222 | 0.0251218 | -0.0248672 | 0.8311500 |
There were no significant relationships with height.
| PC1 | p | PC2 | p | |
|---|---|---|---|---|
| PC1 | -0.0813003 | 0.5746136 | -0.0198157 | 0.8913569 |
There were no significant relationships with weight.
| PC1 | p | PC2 | p | |
|---|---|---|---|---|
| PC1 | -0.0843791 | 0.3807962 | -0.0059876 | 0.9504979 |
There were no significant relationships with GAF.
| PC1 | p | PC2 | p | |
|---|---|---|---|---|
| GAF_C | 0.0454516 | 0.6104405 | -0.1377357 | 0.1210422 |
| GAF_H | 0.0346279 | 0.7014400 | -0.1545339 | 0.0852986 |
There were no significant relationships with MMSE.
| PC1 | p | PC2 | p | |
|---|---|---|---|---|
| PC1 | 0.0536625 | 0.5759074 | -0.235307 | 0.0129165 |
PS=Yes (M = 0.06, SD = 0.26) and PS=No (M = 0.06, SD = 0.26), t(150.00) = 0.00, p > .999, d < 0.01
For the Penn sample (i.e. PS and NC groups), 2 factor scores are computed using the loadings above as weights in a linear combination of relevant gestalt scores (i.e. those gave hits in the 22q sample).
Note: any reference to PCs/factor scores below refers to those computed for the Penn sample.
No significant differences between groups.
##
## Welch Two Sample t-test
##
## data: scores_df$PC1[TD] and scores_df$PC1[PS]
## t = 1.1348, df = 113.37, p-value = 0.2589
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.02576447 0.09485651
## sample estimates:
## mean of TD mean of PS
## 0.3494327 0.3148866
PC2 shows a difference between NC and PS groups.
##
## Welch Two Sample t-test
##
## data: scores_df$PC2[TD] and scores_df$PC2[PS]
## t = -3.8214, df = 109.14, p-value = 0.000221
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.16188349 -0.05131179
## sample estimates:
## mean of TD mean of PS
## -0.1274639 -0.0208663
We see a difference in variances on PC2.
##
## Welch Two Sample t-test
##
## data: scores_df$PC2[SZ] and scores_df$PC2[CR]
## t = 1.4145, df = 52.243, p-value = 0.1631
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.02140075 0.12368864
## sample estimates:
## mean of SZ mean of CR
## -0.005058168 -0.056202110
## Levene's Test for Homogeneity of Variance (center = median)
## Df F value Pr(>F)
## group 1 9.244 0.003667 **
## 53
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
For univariate t-tests, the only difference between groups was in Philtrum measurements:
Height between PS and TD not significantly different.
##
## Welch Two Sample t-test
##
## data: ps_td_height$height[ps_td_height$group == "TD"] and ps_td_height$height[ps_td_height$group == "PS"]
## t = -0.55997, df = 25.921, p-value = 0.5803
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -3.845972 2.199355
## sample estimates:
## mean of x mean of y
## 66.07143 66.89474
Weight between PS and TD not significantly different.
##
## Welch Two Sample t-test
##
## data: ps_td_weight$weight[ps_td_weight$group == "TD"] and ps_td_weight$weight[ps_td_weight$group == "PS"]
## t = -0.63805, df = 58.33, p-value = 0.5259
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -26.93719 13.91420
## sample estimates:
## mean of x mean of y
## 167.6552 174.1667
GAF_C between PS and TD are significantly different.
##
## Welch Two Sample t-test
##
## data: ps_td_gafc$GAF_C[ps_td_gafc$group == "TD"] and ps_td_gafc$GAF_C[ps_td_gafc$group == "PS"]
## t = 7.2342, df = 40.147, p-value = 8.685e-09
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 18.26382 32.42284
## sample estimates:
## mean of x mean of y
## 84.82609 59.48276
GAF_H between PS and TD are significantly different.
##
## Welch Two Sample t-test
##
## data: ps_td_gafh$GAF_H[ps_td_gafh$group == "TD"] and ps_td_gafh$GAF_H[ps_td_gafh$group == "PS"]
## t = 6.6866, df = 38.538, p-value = 6.125e-08
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## 16.43823 30.70420
## sample estimates:
## mean of x mean of y
## 85.26087 61.68966
MMSE between PS and TD are not significantly different.
##
## Welch Two Sample t-test
##
## data: ps_td_mmse$MMSE[ps_td_mmse$group == "TD"] and ps_td_mmse$MMSE[ps_td_mmse$group == "PS"]
## t = 1.9034, df = 47.01, p-value = 0.06313
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
## -0.09303817 3.36133086
## sample estimates:
## mean of x mean of y
## 28.56098 26.92683
## 95% CI: 0.5678-0.7484 (DeLong)
Focusing the prediction on PS subgroups, results are similar to predicting on the aggregated PS group.
## 95% CI: 0.5671-0.7809 (DeLong)
## 95% CI: 0.5053-0.7395 (DeLong)
Prediction strength for PS status drops slightly after adjustment. However, prediction of CR status becomes essentially random post-adjustment.
## 95% CI: 0.5033-0.6931 (DeLong)
## 95% CI: 0.5373-0.7551 (DeLong)
## 95% CI: 0.383-0.6349 (DeLong)
Prediction strength for PS status doesn’t really change after adjusting for height.
## 95% CI: 0.4633-0.845 (DeLong)
Adjusting for height increases prediction of SZ.
## 95% CI: 0.4672-0.89 (DeLong)
## 95% CI: 0.205-0.9665 (DeLong)
Prediction strength for PS status drops slightly after adjusting for weight.
## 95% CI: 0.4624-0.7406 (DeLong)
Adjusting for weight decreases prediction of SZ.
## 95% CI: 0.4807-0.8158 (DeLong)
## 95% CI: 0.3711-0.7151 (DeLong)
Prediction strength for PS status decreases after adjusting for GAF_C.
## 95% CI: 0.3259-0.6516 (DeLong)
Adjusting for GAF_C decreases prediction of SZ.
## 95% CI: 0.3888-0.8431 (DeLong)
Adjusting for GAF_C increases prediction of CR.
## 95% CI: 0.3786-0.7467 (DeLong)
Prediction strength for PS status decreases after adjusting for GAF_H.
## 95% CI: 0.32-0.6455 (DeLong)
Adjusting for GAF_H decreases prediction of SZ.
## 95% CI: 0.4068-0.8541 (DeLong)
Adjusting for GAF_H increases prediction of CR.
## 95% CI: 0.3786-0.7467 (DeLong)
Prediction strength for PS status decreases after adjusting for MMSe.
## 95% CI: 0.4639-0.714 (DeLong)
Adjusting for MMSE decreases prediction of SZ.
## 95% CI: 0.4949-0.7601 (DeLong)
Adjusting for MMSE increases prediction of CR.
## 95% CI: 0.3895-0.7507 (DeLong)
After adjusting PC2 for emotrics, prediction of SZ-status remains strong while CR becomes random.
## 95% CI: 0.5128-0.7028 (DeLong)
## 95% CI: 0.5537-0.7717 (DeLong)
## 95% CI: 0.389-0.6408 (DeLong)